On Fri, 2006-11-03 at 00:05 +0100, Abdelrazak Younes wrote: > Asger Ottar Alstrup wrote: > > - Crashes with multiple windows open. It seems that Buffer contains Rows > > of paragraphs, and this is no good in a multiple view setting. Options: > > a) Rework Buffer to not have Rows (big change, risky), b) Disable > > multiple views completely for now, c) Disable multiple views of same > > document only. Seems to me option C would be safe and still provide some > > benefit to the user. > > Option d) disable multi-window and implement multiple WorkArea within a > QTabWidget. This will allow to open two tabs showing different part of > the same document. This is a nice work-around of the bad "Paragraph row" > model as the two BufferView will have a fortiori the same dimensions. > > > > > > - Cursor trouble: Full refresh on each blink. Abdel says it's a lot of > > work to fix this without the pixamp. Options a) Spend that time, b) > > Revert to old painting scheme. Given that the removal of the pixmap did > > not give us any noticable speed-up, I think we should revert this change > > which was done the last Monday of the meeting. > > I have now restored the backing pixmap. "-dbg painting" will print at > the console which area is refreshed.
Yes, this is a very good idea.
Note that we still are not where we should be (and where 1.4 is): still
after every keypress, the whole screen will be refreshed in the second
update step.
The culprit appears to be the statement
view()->buffer()->changed()
in lyxfunc.C. This sends a signal LyXView to redraw the whole screen in
WorkArea. There is no way to specify anything less than a full screen.
Unfortunately commenting out this line isn't good either: then not even
the current row gets updated (but it will be if you cover and expose
theLyX window...)
This should be redone somehow to do a current-paragraph update if
Update::SinglePar is set. Can you do that with signal-slot?
- Martin
...
> > Other problems without hope:
> >
> > - Mac is unusable because of poor performance, among other things.
I believe it is precisely this issue.
> > Options: a) Revert commit from 14th of July which disabled partial
> > refresh.
>
> I am afraid this is not possible. The BufferView class has changed a
> _lot_ since that time.
Literally, you're correct. Materially something like this is necessary.
> > This is a little dangerous, since it is known that this might
> > cause invalid coord cache entries. b) Do a), but also implement partial
> > coord cache refresh to make it safe. c) Implement a caching painting
> > scheme, such that only changed paint requests are done.
Let's cross that bridge when we get to it.
> I won't say this is without hope. This just needs some time to implement
> the correct solution. This is for sure not a show-stopper
OK.
- Martin
signature.asc
Description: This is a digitally signed message part
